That’s for 65536 byte pieces. Not on zfs in my setup, but on SHR (basically lvm’ed together mdraid 6’s).
This is what my total looks like:
I don’t know why this would happen, but some observations.
- 65536 = 2^16, which is not unusual to see, but you would expect the original segment size to have these common numbers, not really the piece size. The segment size would be 29 times that. Could there be something inherently inefficient at the node end when pieces are this size? Seems unlikely.
- Most of these look like they come from the same satellite, maybe it’s just one customers triggering this behavior
- If all nodes are seeing most of these pieces canceled, that almost necessarily means the customer has actually canceled the whole transfer. Need more sources to confirm.
That’s all I got atm.